JP7253079B2 - Methods for Forming Localization Layers of Digital Localization Maps for Autonomous Driving - Google Patents

Description

The present invention relates to a method for forming a localization layer of a digital localization map for automated driving. The invention further relates to a device for forming a localization layer of a digital localization map for automated driving. The invention further relates to a computer program product.

Mapping technology is known and used to create a map of the surroundings based on sensor observations of mapping vehicles in any number of mapping fleets. Applications of surrounding maps are known, for example, in the fields of robotics, logistics, automotive, aerospace, consumer goods.

Technologies that enable automated driving are generally subject to very high expectations and demands. One of these demands is the availability of finely delineated, accurate, up-to-date and complete digital maps of the surroundings (“digital localization maps”), which are used in AD (automatic driving) or ADAS (English). (advance driver assistance system) is the availability of a digital map of the surroundings (“digital localization map”) that is sufficiently accurate to ensure robust, accurate and reliable positioning, planning and navigation of vehicles.

Today, automated vehicles rely on map-based information that assists vehicles in route planning, object recognition, and situational estimation. These maps are usually called HD maps as opposed to SD maps used for traditional navigation applications.

In such high-resolution digital localization maps, three layers are generally distinguished:

a. A planning layer (at the bottom) containing e.g. precise lane geometry and connectivity information. Planning layers are used, among other things, to keep the automated vehicle in its lane and to assist the vehicle in maneuver planning.

b. A (middle) localization layer containing objects (“landmarks”) detected by different sensor types (e.g. radar sensors, lidar sensors, cameras, etc.). The instantaneous information captured by each sensor is matched with localization layers for the purpose of determining the vehicle's position within the digital localization map or surroundings.

c. A (top) dynamic layer containing dynamically changing information such as road conditions, traffic conditions, weather, parking space conditions, etc. Dynamic layers are primarily used for comfort functions within digital localization maps.

Because digital localization maps for autonomous driving provide high temporal and spatial positioning accuracy, along with improved accuracy of digital localization maps for autonomous driving compared to conventional localization maps (e.g. for navigation systems) , it is also desirable to ensure up-to-dateness.

Using the localization layer, among other things, feature-based positioning systems can provide localization for mobile systems that sense their surroundings (e.g., automated vehicles) by decoding point coordinates from observable landmarks. can. Such systems recognize landmark elements (eg, landmarks in the form of traffic signs, lane tracks, road markings, etc.) at specific locations in the surroundings. Measurement of the visible angle of the movement mechanism with respect to the landmark element then allows determination of the vehicle position with respect to the landmark.

WO2017/174229 discloses a method for determining the pose of an at least partially autonomous vehicle within a surrounding using landmarks. In this case, the vehicle is moving in the surroundings, thereby generating a series of localization scenarios in which the landmark data is at least digitally processed by the vehicle control system to determine the pose of the vehicle. It is proposed to increase or decrease the amount of landmark data depending on the localization scenario as needed.

DE 102011119762 A1 discloses a system and method for determining the position of a motor vehicle, in which a position determination system suitable for motor vehicles and a corresponding method are proposed. The system includes a digital map that is located and populated with data about location-specific features, at least one environment awareness device for capturing location-specific features of the vehicle's surroundings, a digital map and the surroundings. An environment recognition device and an associated localization module are included. A localization module matches the captured data with data populated in the digital map for location-specific features and locates the vehicle position based on the location-specific features populated within the digital map. It has a processing unit for identifying.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an improved method for creating a localization layer of a digital localization map.

The problem is, according to a first embodiment, a method for forming a localization layer of a digital localization map for automated driving, comprising:
- providing a localization layer for a given area;
- providing a planning layer for this area;
- extracting alignment features from the planning layer and the localization layer for alignment, such that unacceptable deformations of the localization layer can be recognized during alignment of the planning layer and the localization layer; , is resolved by how the alignment features are extracted from the localization layer.

With the localization layer thus generated, the alignment of the planning map and the localization map with each other based on the alignment features of the localization map is such that errors or unacceptable differences between the localization layer and the planning layer can be easily detected. has the advantage of being implemented This has the advantage that a correct and usable digital localization map can be generated.

Advantageously, with the proposed method, it is not necessary to inspect the entire data set in order to inspect the accuracy of the registration accuracy of the localization layer and the planning layer. In this way, among other things, it is efficiently possible to use planning layers and localization layers from different manufacturers in the creation of the digital localization map.

The subject is, according to a second embodiment, a device for forming a localization layer of a digital localization map for automated driving, comprising:
- a provisioning device for providing a localization layer for a given area and for providing a planning layer for this area;
- an extraction device for extracting alignment features from the planning layer and the localization layer for alignment, wherein unacceptable deformations of the localization layer are recognized during alignment of the planning layer and the localization layer; To obtain, the alignment features are resolved by a device that can be extracted from the localization layer.

This problem is solved, according to a third embodiment, by a computer program product.
Advantageous variants of the method are subject matter of the dependent claims.

An advantageous variant of the method contemplates that the registration features are defined structurally. By doing this, the alignment features in the localization layer have a structure that is easy to notice when the above layers are aligned.

A further advantageous variant of the method is arranged such that the alignment features are formed according to a defined pattern. Even with this measure, unacceptable deformations of the localization layer are easily noticed when aligning the above layers.

A further advantageous variant of the method is arranged such that the alignment features are formed on the basis of at least one defined statistical measure. This allows the introduction of sophisticated statistical tools for extracting and forming alignment features, by means of which unacceptable differences in the alignment of the above layers can be easily recognized.

A further advantageous variant of the method is characterized in that the statistical measure comprises density distribution and/or variance. This provides a known statistical measure for recognizing the deformation of registration features during registration of the above layers.

A further advantageous variant of the method is characterized in that the maximum permissible degree of distortion of the registration features during registration of the localization layer and the planning layer is defined by a threshold value. In this way, it is advantageous to provide some sort of check quantity that can define how much more deformation of the check layer is tolerable. This provides a threshold that defines how far the alignment of the above layers is acceptable to provide a usable digital localization map for autonomous driving. This is particularly advantageous if localization and planning layers from different manufacturers are provided or if both said layers are from different sensor sets.

In the following the invention is explained in greater detail with further features and advantages on the basis of several figures. In this regard, the technical features and advantages of the aforementioned device are evident from the technical features and device of the aforementioned method.

1 is a principle diagram of the proposed device for forming a localization layer of a digital localization map for automated driving; FIG. FIG. 12 is a diagram for the rationale of the proposed placement of the registration features of the localization layer; FIG. 12 is a diagram for the rationale of the proposed placement of the registration features of the localization layer; FIG. 12 is a diagram for the rationale of the proposed placement of the registration features of the localization layer; 1 is a principle diagram of a flow chart of a proposed method for forming a localization layer of a digital localization map for automated driving; FIG.

A central idea of the invention is, inter alia, to allow an improved registration of the localization layer and the planning layer of a digital localization map. This makes it possible, inter alia, to recognize errors in the alignment of the layers mentioned above and thus to efficiently check the creation process of the digital localization map.

The alignment process preferably applies a non-linear transformation to the localization layer to precisely align the localization layer to the planning layer. It must be ensured that the applied transformations do not destroy or corrupt the digital localization map (error checking). In this regard, statistical measures are preferably performed before and after the alignment process to determine if the alignment process is unacceptable or if the alignment cannot be successfully performed.

If the alignment features are derived such that they have significant statistical properties or parameters (e.g. density distribution, variance, etc.) of the localization layer from which they are derived, then the alignment features themselves can be transformed based on non-linear transformations. It can be used to estimate the errors introduced.

By doing so, it is advantageous not to have to use the complete data material of the above layers to check the registration process. This serves data protection interests, for example, if the map layers mentioned above are from different manufacturers who have data protection interests in respect of the provided map layers.

With the proposed method, the computation of periodic measures based on this quantitatively reduced registration feature is generally much faster than using the full data of the localization layer, thus accelerating the registration process. It is advantageous to obtain

In the localization layer registration process, the localization layer is subjected to a non-linear transformation whose goal is to establish the identified registration features in both layers (localization layer and planning layer) with sufficient accuracy. and Since the identification of matching alignment objects or alignment features is an automated process, errors can occur in doing so (e.g., incorrect localization layers can render the localization layer useless to the localization process from time to time). displacement).

Therefore, according to the invention, statistical measures, such as density distributions or variances, are determined before and after the non-linear transformations described above for the extraction of registration features. Analysis of this statistical measure can be used to determine erroneous transformations (eg, if the difference in density of alignment features or localization objects within the post-transformation virtual region exceeds a threshold).

Alignment Features 1a. . . It is proposed to compute registration features in such a way that error checking can be performed directly using 1n, without the need to use a full localization layer for that purpose. For this purpose, registration features are chosen such that errors are easily recognizable.

The following figure qualitatively illustrates the proposed principle of registration feature extraction.
FIG. 1 shows a block diagram of an apparatus 100 for forming a localization layer of a digital localization map for automated driving. The apparatus 100 has a providing device 10 for providing a localization layer 12 for a defined area and for providing a planning layer 11 for this area. Apparatus 100 extracts alignment features 1a . from localization layer 12 for alignment with planning layer 11 . . . 1n, and in this case the alignment features 1a.1n, so that impermissible deformations of the localization layer 12 can be recognized during the alignment of the planning layer 11 and the localization layer 12. . . 1n can be extracted from the localization layer 12.

FIG. 2 shows a principle diagram of a localization layer 12 with landmarks captured using, for example, radar sensors.

FIG. 3 illustrates conventional alignment features 1a.1 extracted from the localization layer 12 of FIG. . . A localization layer 12 with 1d is shown. In this case, alignment objects 1a . . . The 1d statistical distribution does not correspond to the landmark placement in FIG.

In FIG. 4, registration feature 1a. . . 1n are shown arranged here in a structured manner according to the invention, in the case of FIG. The statistical distribution of matching objects corresponds to the placement of landmarks in FIG.

Alignment feature 1a. of FIG. . . 1n, the registration features of the localization layer 12 to the planning layer 11 (not shown) by way of illustration only and by way of example shown regular placement. . . It can be seen that the unacceptable distortion of 1n can be easily recognized, thereby pinpointing the erroneous alignment of the localization layer 12 and the planning layer 11 . As a result, the correct creation process of digital localization maps is already assisted early on.

It is an advantage that the surrounding environment data provided by the method according to the invention can produce a digital localization map that is very accurate, robust, as complete as possible and highly available. Advantageously, the general principle is very simple and enables many implementations and applications described above merely by way of example.

Apparatus 100 is preferably configured as a computing device in a cloud on which computer program products are executed.

Figure 5 shows the principle flow of the proposed method for forming the localization layer of a digital localization map for automated driving.

At step 200, provision of a localization layer for a defined region is performed.

At step 210, provision of a planning layer for this area is performed.

Finally, in step 220, the alignment features 10a. from the localization layer for alignment with the planning layer. . . 10n is performed, wherein alignment features 10a . . . 10n are extracted from the localization layer.

Advantageously, the method according to the invention may be implemented as software, preferably running on a cloud-based server device. This supports easy adaptability of the method.

Those skilled in the art will modify and/or combine features of the present invention with each other as appropriate without departing from the spirit of the invention.

Claims (6)

A method for forming a localization layer (12) of a digital localization map for automated driving, comprising:
- providing said localization layer (12) for a defined region;
- providing a planning layer (11) for said area;
- extracting alignment features (1a...1n) from said localization layer (12) for alignment with said planning layer (11), wherein said planning layer (11) and said said alignment features (1a...1n) are extracted from said localization layer (12) such that unacceptable deformations of said localization layer (12) can be recognized during alignment of said localization layer (12) ;
A method , wherein said alignment features (1a...1n) are formed according to a defined pattern . Method according to claim 1, wherein said registration features (1a...1n) are formed according to at least one defined statistical measure. 3. The method of claim 2 , wherein the statistical measure comprises density distribution and/or variance. 2. A threshold defines the maximum allowable degree of distortion of the registration features (1a...1n) during registration of the localization layer (12) and the planning layer (11). 4. The method according to any one of 1 to 3 . An apparatus (100) for forming a localization layer (12) of a digital localization map for automated driving, comprising:
- a providing device (10) for providing said localization layer (12) for a defined area and for providing a planning layer (11) for said area;
- an extraction device (120) for extracting alignment features (1a...1n) from said planning layer (11) and said localization layer (12) for alignment, wherein said planning Said alignment features (1a...1n) are defined by said localization layer ( 12) can be extracted from
An apparatus (100) wherein said alignment features (1a...1n) are formed according to a defined pattern . Any of claims 1 to 4 when operating on an electronic device (100) or stored on a computer readable data storage medium for forming a localization layer for a digital localization map for automated driving. A computer program product comprising program code means for implementing the method of claim 1.

Images